This invention relates to an echo cancelling device and, in particular, to an echo cancelling device using a neural network.
Recently, an echo cancelling device has been used in various types of communication equipments or systems, such as a teleconference system, or the like.
An example of such a conventional echo cancelling device is disclosed in Japanese Unexamined Patent Publication No. 8755/1989 or No. 83491/1991.
For example, the Japanese Unexamined Patent Publication No. 8755/1989 shows a teleconference system using an echo cancelling device. The teleconference system comprises a microphone, a loudspeaker, an echo canceller, a level detection circuit, a comparator, a level detection circuit, a switch, a noise producing circuit, a communication circuit, and two amplifier circuits. One of the amplifier circuits is supplied as an input transmission speech signal with a speech signal produced by the microphone and amplifies the input transmission speech signal to produce an amplified transmission speech signal. The amplified transmission speech signal is sent through a transmission path to be led to the communication circuit as an output transmission speech signal. The communication circuit delivers the output transmission speech signal to a channel. On the other hand, the communication circuit receives an input reception speech signal through the channel and delivers the input reception speech signal through a reception path to another one of the amplifier circuits. The another one of the amplifier circuits amplifies the input reception speech signal to produce an amplified reception speech signal which is supplied to the loudspeaker as an output reception speech signal. The loudspeaker converts the output reception speech signal into a reception speech sound. The reception speech sound intrudes or enters into the microphone as an echo signal. The echo canceller for cancelling such intrusion of the reception speech sound is connected between the reception path and the transmission path.
However, in the above-described teleconference system, every time when an acoustic echo condition around the loudspeaker and the microphone is suddenly deteriorated (for example, the teleconference system is located in a mobile compartment and the door of the compartment is opened and closed), the loudspeaker produces the white noise for a training operation. This interferes the communication. Even if the acoustic echo condition is not so deteriorated, an adaptive calculation process is required in order to conform to the variation of the condition. Such calculation process takes a time duration ranging from several hundred milliseconds to several seconds. In this time duration, an unstable state is continued where a strong echo sound is audibly felt.
On the other hand, an echo cancelling device disclosed in the Japanese Unexamined Patent Publication No. 83491/1991 comprises a howling detecting section for detecting a communication fault such as howling, a training sound producing section for producing a training sound when the howling detecting section detects the communication fault, and an echo learning section for resetting a new fundamental echo characteristic when the training sound producing section produces the training sound. In the above-mentioned echo cancelling device also, the training sound is produced every time when the howling detecting section detects occurrence of howling. In that event, the communication is interfered like in the above-mentioned teleconference system.